Seasonal Residential Water Demand Forecasting for Census Tracts

Abstract

The paucity of readily available demographic, economic, and water consumption data at household levels has limited the application of disaggregate water demand models. This research develops regression-based water demand models capable of forecasting single-family residential water demands within individual census tracts at a bimonthly time-step. The regression models are estimated using 12 years of demographic, weather, economic, and metered bimonthly water consumption data associated with over 100 unique census tracts in Seattle, Washington. In general, the three regression methods perform well in replicating total single-family water consumption in the study region. Two regression models, a fixed effects model and a random effects model, provide better estimates of water demand within individual census tracts. Improved water demand forecasts at the spatial scale of census tracts provide policy makers and planners information useful for managing water resources. These proposed approaches allow examination of spatially distributed demands within systems, identify the value of targeted conservation and infrastructure development, and improve understanding of the variables impacting demand in heterogeneous areas. The coefficient estimates developed in this research are appropriate for use in spatially disaggregate urban simulation models.